Hamed Zamani Receives 2025 Amazon Research Award to Advance Efficient Agentic AI
Content
Associate Professor Hamed Zamani of the Manning College of Information and Computer Sciences (CICS) has received a 2025 Amazon Research Award to support his work on making artificial intelligence systems more efficient and cost-effective. This is Zamani’s second Amazon Research Award; he previously received the award in 2022 for work on retrieval-enhanced machine learning models. Backed by $70,000 from Amazon, Zamani’s project, “A Framework for Proactive and Collaborative AI Agents,” focuses on a growing challenge in artificial intelligence: how to make AI agents powerful enough to complete complex tasks, but efficient enough to use in practical settings. His Cost-Aware and Proactive (CAP) agent framework is designed to help agents collaborate, reuse workflows, and account for cost when determining how to complete a task.
As generative AI systems increasingly leverage external tools and collaborate with other AI agents, they are opening new possibilities for solving complex problems. But these systems can also be expensive to run, slow to respond, and largely reactive, waiting for user prompts rather than anticipating useful next steps.
Zamani’s CAP framework is designed to address these challenges by allowing less expensive agents to draw on existing workflows, rather than requiring a more powerful agent to handle every task from scratch.
“In e-commerce,” Zamani said, “such technologies have the potential to power the next generation of shopping assistants—agents that not only search for individual products, but also identify bundled offers, optimize the use of coupons and discounts, analyze reviews, and deliver a personalized, well-explained experience to help customers make informed decisions.”
The system is inspired by "blackboard" architecture that was pioneered at CICS by Distinguished Professor Emeritus Victor Lesser, a design in which multiple agents read from and write to a shared workspace. Using that shared information, agents can proactively “propose” actions, with each proposal including an estimated cost. These estimates allow tasks to be allocated to agents based on cost and the requirements of the task.
“We believe the proposed research will significantly enhance our understanding of the agentic AI landscape and advance the state of the art in this field,” Zamani said.
Zamani, who joined the Manning CICS faculty in 2020, is a recipient of the U.S. National Science Foundation CAREER Award as well as the ACM SIGIR Early Career Excellence in Research and Community Engagement Awards. He holds a doctorate in computer science from the University of Massachusetts Amherst.
